CN108961460B - Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization - Google Patents
Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization Download PDFInfo
- Publication number
- CN108961460B CN108961460B CN201810789980.6A CN201810789980A CN108961460B CN 108961460 B CN108961460 B CN 108961460B CN 201810789980 A CN201810789980 A CN 201810789980A CN 108961460 B CN108961460 B CN 108961460B
- Authority
- CN
- China
- Prior art keywords
- prediction
- model
- sparse
- service life
- objective
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G07—CHECKING-DEVICES
- G07C—TIME OR ATTENDANCE REGISTERS; REGISTERING OR INDICATING THE WORKING OF MACHINES; GENERATING RANDOM NUMBERS; VOTING OR LOTTERY APPARATUS; ARRANGEMENTS, SYSTEMS OR APPARATUS FOR CHECKING NOT PROVIDED FOR ELSEWHERE
- G07C3/00—Registering or indicating the condition or the working of machines or other apparatus, other than vehicles
- G07C3/005—Registering or indicating the condition or the working of machines or other apparatus, other than vehicles during manufacturing process
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/20—Administration of product repair or maintenance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/04—Manufacturing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2111/00—Details relating to CAD techniques
- G06F2111/06—Multi-objective optimisation, e.g. Pareto optimisation using simulated annealing [SA], ant colony algorithms or genetic algorithms [GA]
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention discloses a fault prediction method and a fault prediction device based on sparse ESGP and multi-objective optimization, wherein the method comprises the following steps: collecting operation data of the equipment to be tested, and extracting trend characteristics of the equipment to be tested according to the operation data; introducing a sparse Gaussian process to improve the Gaussian process of the echo state Gaussian process, further introducing a multi-objective genetic algorithm to perform parameter optimization on a model of the echo state Gaussian process, and simultaneously taking the point prediction and interval prediction effects of the remaining service life as the objective of multi-objective optimization; according to the method, the complex model parameters are automatically optimized through multi-objective optimization, a better ESGP prediction model is selected more efficiently, and the remaining service life of the equipment to be tested is finally predicted through the optimized model according to the trend characteristics so as to obtain a better prediction result, so that the downtime and maintenance cost of the equipment can be effectively reduced, the operation efficiency and safety of the equipment are improved, and the method has an industrial application value.
Description
Technical Field
The invention relates to the technical field of fault prediction, in particular to a fault prediction method and a fault prediction device based on sparse ESGP (Echo State Gaussian Processes) and multi-objective optimization.
Background
At present, with the development of industrial 4.0 and intelligent manufacturing, a Predictive Maintenance (Predictive Maintenance) technology has important significance in the aspects of improving the Maintenance quality of equipment, reducing the Maintenance cost, reducing the running risk of the equipment and the like. Predictive maintenance is an important objective of a PHM (fault prediction and Health Management) system, and one of the core contents for implementing predictive maintenance is to predict the Remaining service Life of the equipment, namely RUL (Remaining Useful Life). Reliable remaining life prediction can provide meaningful information for decision making for equipment maintenance, thereby avoiding catastrophic failure of the system.
In the related art, the failure prediction method is mainly divided into three categories: model-based, data-driven, model-and-data-driven combined methods. With the development of technologies such as sensors and storage, the big data background makes the data-driven fault prediction method to be of great interest. The existing fault prediction method based on data driving is mainly divided into two categories, namely a statistical model-based method and a machine learning-based method, and aims to learn trend characteristics from a large amount of data so as to discover the behavior of a system in the operation process. Common methods include artificial neural networks, support vector regression, Echo state networks (Echo state networks), and the like, and some recently popular deep learning methods such as LSTM (Long Short-Term Memory network), DBN (deep belief networks), and the like.
However, most of related art fault prediction methods based on data driving are limited to point prediction of RUL, and prediction uncertainty cannot be considered, so that the obtained RUL prediction result information is single, and decision activities of subsequent health management or predictive maintenance cannot be strongly supported.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, one purpose of the invention is to provide a fault prediction method based on sparse ESGP and multi-objective optimization, which can effectively reduce the downtime and maintenance cost of equipment, improve the operation efficiency and safety of the equipment and has industrial application value.
The invention also aims to provide a fault prediction device based on sparse ESGP and multi-objective optimization.
In order to achieve the above object, an embodiment of the present invention provides a fault prediction method based on sparse ESGP and multi-objective optimization, including the following steps: collecting operation data of equipment to be tested, and extracting trend characteristics of the equipment to be tested according to the operation data; introducing a sparse Gaussian process to improve the Gaussian process of the echo state Gaussian process, introducing a multi-objective genetic algorithm to perform parameter optimization on a model of the echo state Gaussian process, and taking the effects of point prediction and interval prediction of the residual service life as the target of multi-objective optimization; and finally predicting the residual service life of the equipment to be tested through the optimized model according to the trend characteristics so as to obtain a better prediction result.
The fault prediction method based on sparse ESGP and multi-objective optimization mainly predicts the residual service life of equipment or key structural parts of the equipment, the introduction of sparse GP is suitable for large-scale data, and complicated model parameters are automatically optimized through multi-objective optimization, so that a better ESGP prediction model is more efficiently selected, reliable RUL prediction with prediction intervals is provided under the consideration of the prediction uncertainty, valuable information is provided for predictive maintenance of the equipment, the downtime and maintenance cost of the equipment can be effectively reduced, the operation efficiency and safety of the equipment are improved, and the fault prediction method has industrial application value.
In addition, the fault prediction method based on sparse ESGP and multi-objective optimization according to the above embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the extracting a trend characteristic of the device under test according to the operation data further includes: acquiring expressed multidimensional feature data acquired by a plurality of sensors, and selecting to remove redundant features according to statistical indexes; and extracting the characteristics of each dimension of original characteristics, setting the length of a time window as a preset value, and taking the average value and the difference value of each time window as the extracted characteristics so as to obtain the trend characteristics of the equipment to be tested.
Further, in an embodiment of the present invention, the method further includes: establishing a mathematical model of the echo state Gaussian process, using the trend characteristics as model input and the residual service life as model output, solving an output weight matrix according to an update equation of the echo state network, and keeping the iteratively updated reserve pool state; and combining the model input and the updated reserve pool state to obtain the input of a sparse Gaussian process regression model, and training by taking the true value of the residual service life as the target output of the sparse Gaussian process regression model.
Further, in an embodiment of the present invention, the method further includes: before training, selecting a linear function as a mean function and selecting a square exponential function as a variance function; and taking the preprocessed test data as input to test to obtain a prediction result of the residual service life after one iteration.
Further, in an embodiment of the present invention, the introducing a sparse gaussian process to improve a gaussian process of the echo state gaussian process, and introducing a multi-objective genetic algorithm to perform parameter optimization on a model of the echo state gaussian process, and taking an effect of point prediction and interval prediction as an objective of the multi-objective optimization, further includes: taking the average value of the prediction result as the point prediction of the residual service life, and using the variance for constructing a prediction interval to obtain a measurement index of the prediction result; obtaining a new comprehensive index according to the measurement indexes of point prediction and interval prediction in the prediction result, and constructing an optimization objective function by combining the new comprehensive index, so that a multi-objective problem is converted into a single-objective problem, and the optimization is performed by using a genetic algorithm, wherein in the optimization process of the genetic algorithm, a cross validation method with weight is introduced and the fitness is calculated by combining the objective function constructed in the front, so that a better chromosome is selected through iterative optimization, the chromosome is a model parameter to be optimized, and an optimized sparse ESGP model is provided by combining the optimized model parameter; in the testing stage, training data and testing data are simultaneously used as input, the likelihood function is selected as Gaussian likelihood, the inference mode is selected as Gaussian inference, and the testing is carried out based on the trained sparse ESGP model so as to obtain the predicted value of the residual service life and the mean value and the variance of the residual service life.
In order to achieve the above object, an embodiment of another aspect of the present invention provides a fault prediction apparatus based on sparse ESGP and multi-objective optimization, including: the acquisition device is used for acquiring the operation data of the equipment to be detected and extracting the trend characteristics of the equipment to be detected according to the operation data; the introduction module is used for introducing a sparse Gaussian process to improve the Gaussian process of the echo state Gaussian process, introducing a multi-objective genetic algorithm to perform parameter optimization on a model of the echo state Gaussian process, and taking the effects of point prediction and interval prediction of the residual service life as the target of multi-objective optimization; and the prediction module is used for finally predicting the residual service life of the equipment to be tested through the optimized model according to the trend characteristics so as to obtain a better prediction result.
The fault prediction device based on sparse ESGP and multi-objective optimization mainly predicts the residual service life of equipment or key structural parts of the equipment, the introduction of sparse GP is suitable for large-scale data, and complicated model parameters are automatically optimized through multi-objective optimization, so that a better ESGP prediction model is more efficiently selected, reliable RUL prediction with prediction intervals is provided under the consideration of the prediction uncertainty, valuable information is provided for predictive maintenance of the equipment, the downtime and maintenance cost of the equipment can be effectively reduced, the operation efficiency and safety of the equipment are improved, and the fault prediction device has industrial application value.
In addition, the fault prediction device based on sparse ESGP and multi-objective optimization according to the above embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the acquisition module is further configured to acquire multi-dimensional feature data acquired by a plurality of sensors, select to eliminate redundant features according to statistical indexes, perform feature extraction on each dimension of original features, set a length of a time window as a preset value, and take an average value and a differential value of each time window as extracted features to obtain a trend feature of the device under test.
Further, in an embodiment of the present invention, the method further includes: and the modeling module is used for establishing a mathematical model of the echo state Gaussian process, using the trend characteristics as model input and the residual service life as model output, solving an output weight matrix according to an update equation of the echo state network, reserving the iteratively updated reserve pool state, combining the model input and the updated reserve pool state to obtain the input of a sparse Gaussian process regression model, and training by using the true value of the residual service life as the target output of the sparse Gaussian process regression model.
Further, in an embodiment of the present invention, the method further includes: and the calculation module is used for selecting a linear function as a mean function and a square exponential function as a variance function before training, and taking the preprocessed test data as input to test to obtain a prediction result of the residual service life after one iteration.
Further, in an embodiment of the present invention, the introducing module is further configured to use a mean value of the prediction result as a point prediction of the remaining service life, and use the variance for constructing a prediction interval to obtain a metric index of the prediction result, obtain a new comprehensive index according to the metric index of the point prediction and the interval prediction in the prediction result, and construct an optimization objective function in combination with the new comprehensive index, thereby converting the multi-objective problem into a single-objective problem, so as to perform optimization with a genetic algorithm, wherein in an optimization process of the genetic algorithm, a cross validation method with weight is introduced and a fitness is calculated in combination with a previously constructed objective function, so as to select a superior chromosome, which is a model parameter to be optimized, by iterative optimization, and further give an optimized sparse ESGP model in combination with the optimized model parameter, and in the testing stage, simultaneously taking training data and testing data as input, selecting a likelihood function as Gaussian likelihood, selecting an inference mode as Gaussian inference, and testing based on the trained sparse ESGP model to obtain a predicted value of the residual service life and a mean value and a variance of the residual service life.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
FIG. 1 is a flow diagram of a sparse ESGP and multi-objective optimization based fault prediction method according to one embodiment of the invention;
FIG. 2 is a flow chart of a turbine engine remaining useful life prediction based on sparse ESGP, according to an embodiment of the present invention;
FIG. 3 is a block diagram of a sparse ESGP method according to one embodiment of the present invention;
FIG. 4 is a flow diagram of the optimization of sparse ESGP model parameters based on a multi-objective genetic algorithm, according to one embodiment of the present invention;
FIG. 5 is a graph of results of predictions of remaining useful life for 100 turbine engines based on the sparse ESGP method according to one embodiment of the present invention;
fig. 6 is a schematic structural diagram of a fault prediction device based on sparse ESGP and multi-objective optimization according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The following describes a fault prediction method and a fault prediction device based on sparse ESGP and multi-objective optimization according to an embodiment of the present invention with reference to the drawings, and first, a fault prediction method based on sparse ESGP and multi-objective optimization according to an embodiment of the present invention will be described with reference to the drawings.
Fig. 1 is a flowchart of a fault prediction method based on sparse ESGP and multi-objective optimization according to an embodiment of the present invention.
As shown in fig. 1, the fault prediction method based on sparse ESGP and multi-objective optimization includes the following steps:
in step S101, operation data of the device to be tested is collected, and a trend feature of the device to be tested is extracted according to the operation data.
In an embodiment of the present invention, extracting a trend characteristic of the device under test according to the operation data further includes: acquiring expressed multidimensional feature data acquired by a plurality of sensors, and selecting to remove redundant features according to statistical indexes; and extracting the characteristics of each dimension of original characteristics, setting the length of a time window as a preset value, and taking the average value and the difference value of each time window as the extracted characteristics so as to obtain the trend characteristics of the equipment to be tested.
Specifically, as shown in fig. 2, the present invention is implemented by first performing data preprocessing, including: (1) selecting characteristics: based on the multi-dimensional feature data acquired by a plurality of sensors, the data features are selected by using statistical indexes such as variance and correlation coefficient, specific threshold values are set, and features with over-small variance and over-large correlation coefficient are removed to reduce redundant features and reduce calculated amount. (2) Feature extraction: and for each dimension of original feature extraction trend features, setting the length of a time window to be L, and taking the average value and the difference value of each given time window as the extracted features. Noise can be reduced to a certain extent, and the reduction of the data volume can improve the computational efficiency of the method.
In step S102, a sparse gaussian process is introduced to improve the gaussian process of the echo state gaussian process, so as to alleviate the problem of high computational complexity of the general gaussian process, and a multi-objective genetic algorithm is introduced to perform parameter optimization on the model of the echo state gaussian process, and meanwhile, the effects of point prediction and interval prediction of the remaining service life are used as the objective of multi-objective optimization.
It can be understood that, in the embodiment of the present invention, the GP portion of the original ESGP Algorithm is improved, a sparse gaussian process is introduced, the problem of high computational complexity of a general gaussian process is alleviated, so as to be suitable for processing data of a larger scale, an MOGA (Multi-object Genetic Algorithm) is introduced to perform parameter optimization on the ESGP model, and meanwhile, the effects of point prediction and interval prediction are considered as the target of Multi-target optimization, so that automatic optimization of model parameters is performed while the effects of point prediction and interval prediction are improved.
Further, in an embodiment of the present invention, the method further includes: establishing a mathematical model of the echo state Gaussian process, using the trend characteristics as model input and the residual service life as model output, solving an output weight matrix according to an update equation of an echo state network, and keeping the iteratively updated reserve pool state; and combining the model input and the updated reservoir state to obtain the input of the sparse Gaussian process regression model, and training by taking the true value of the residual service life as the target output of the sparse Gaussian process regression model.
In one embodiment of the present invention, further comprising: before training, selecting a linear function as a mean function and selecting a square exponential function as a variance function; and taking the preprocessed test data as input to test to obtain a prediction result of the residual service life after one iteration.
For example, as shown in fig. 2 and 3, the embodiment of the present invention is described with specific application to RUL prediction of a turbine engine, specifically:
(1) ESN training stage: establishing an ESGP mathematical model of the turbine engine, using the selected and extracted features in A as model inputs u (t), and the true RUL as model output ytarget(t) obtaining an output weight matrix W according to an ESN update equation and a ridge regression principleoutIn addition, iteratively updated reserve pool state h (t) is retained.
(2) Sparse ESGP training: merging the original model input u (t) with the updated reserve pool state h (t) to obtain
And will be
Truth y of RUL as input to the sparse Gaussian Process regression modeltarget(t) purpose as modelAnd (4) training the standard output, wherein the training process comprises the determination of a mean function and a variance function, the determination of an inference function and the like.
(3) Sparse gaussian process: before training, the mean function and variance function of GP are determined, linear function is selected as the mean function, and common square exponential function is selected as the variance function, namely:
the inference process selects FITC (Fully Independent Training condition) for approximate inference.
(4) Sparse ESGP test: and (4) taking the test data after the pretreatment of the A as input, and testing to obtain an RUL prediction result after one iteration.
Further, in an embodiment of the present invention, introducing a sparse gaussian process to improve a gaussian process of the echo state gaussian process, and introducing a multi-objective genetic algorithm to perform parameter optimization on a model of the echo state gaussian process, and taking an effect of point prediction and interval prediction as an objective of the multi-objective optimization, further includes: taking the average value of the prediction result as the point prediction of the residual service life, and using the variance for constructing a prediction interval to obtain the measurement index of the prediction result; and (3) obtaining a new comprehensive index according to the measurement indexes of point prediction and interval prediction in the prediction result, and combining the new comprehensive index to construct an optimization objective function, so that the multi-objective problem is converted into a single-objective problem, and the optimization is performed by using a genetic algorithm. The chromosome is a model parameter which is required to be optimized, and an optimized sparse ESGP model is further provided by combining the optimized model parameter.
Specifically, the embodiment of the invention carries out model multi-objective optimization, including;
(1) metric index of prediction result: the mean of the RUL prediction results can be directly used as the point prediction for RUL, and the variance can be used to construct the prediction interval. The measure of the effectiveness of point prediction can be expressed by Mean Square Error (MSE) and a scoring function S, i.e.:
wherein RULi、RULtrueThe predicted and true values of RUL, diIs an error term and has di＝RULi-RULtrue。
On the other hand, given a confidence of 95%, a prediction interval is constructed using variance information, and the quality of the prediction interval is evaluated using the prediction interval coverage PICP and the standard average interval width NMPIW, i.e.:
where n is the number of samples, I (-) represents the indicator function when
(i.e., the true value of RUL is within the interval bounded by the upper and lower bounds) I (· 1), otherwise I (· 0) and Ω ═ RULmax-RULmin
(2) Constructing and optimizing objective function, 1) combining two measurement indexes PICP and NMPIW of prediction interval to provide new comprehensive index CWCnewNamely, the following steps are provided:
wherein gamma (PICP) ═ 0, when PICP ≥ mu1Otherwise γ (PICP) ═ 1, and γ (NMPIW) is similar. η1、η2、μ1And mu2Two hyper-parameters that measure the degree of penalty are controlled. CWC (continuous wave conductor)newThe quality of the PICP and the NMPIW can be respectively controlled to a certain extent, so that the comprehensive quality of a prediction interval is evaluated.
2) In addition, a new objective function is further provided by combining the point prediction error MSE of the RUL, and a multi-objective problem is converted into a single-objective problem in a weighted fusion mode, namely:
wherein
Representing the weight occupied by the mean square error term MSE, η3And mu3The penalty level of MSE is controlled for the constant parameter.
(3) Multi-objective genetic algorithm (MOGA) hyper-parameter optimization: the hyper-parameters in the sparse ESGP model mainly comprise two parts, namely structure parameters of ESN and model parameters of GP part, such as variance of noise signals, likelihood function, mean function and other related parameters. The former is mainly determined by introducing a cross validation method with weight, and the latter is mainly determined by a conjugate gradient descent method according to the maximum likelihood principle. And optimizing the parameters of the ESN part and the parameters of the GP part by utilizing MOGA. The optimization objective is L (MSE, CWC) in equation 7new) The specific flow of MOGA is shown in FIG. 4.
In step S103, the remaining service life of the device to be tested is finally predicted through the optimized model according to the trend characteristics, so as to obtain a better prediction result.
In one embodiment of the invention, in the testing stage, training data and testing data are simultaneously used as input, the likelihood function is selected as Gaussian likelihood, the inference mode is selected as Gaussian inference, and the testing is carried out based on a trained sparse model to obtain a predicted value of the residual service life and the mean value and the variance of the residual service life.
Specifically, RUL predicts: and in the testing stage, the training data and the testing data are simultaneously used as input, the likelihood function is selected as Gaussian likelihood, the inference mode is selected as Gaussian inference, the testing is carried out based on the previously trained sparse ESGP model, the predicted value of the residual service life of the turbine engine, the mean value and the variance of the RUL are obtained, and the RUL prediction is completed.
In addition, as shown in fig. 5, the embodiment of the present invention is a result graph of predicting the remaining service life of 100 turbine engines (single failure mode, under the condition of constant operating conditions) based on the sparse ESGP method; specific metrics are Score 347, MSE 250, PICP 1 and NMPIW 0.896.
In summary, the original ESGP method is a method based on the fusion of the Echo State Network (ESN) and the Gaussian Process (GP), and can give the RUL prediction with prediction interval, but there are two problems: 1) the ESGP model is relatively complex, the number of related key parameters is large, and training is difficult; 2) due to the influence of GP, ESGP has high calculation complexity, low speed when processing large-scale data and influences real-time effect. The method provided by the embodiment of the invention mainly utilizes the echo state Gaussian process to predict the residual service life, simultaneously gives the prediction interval of the RUL to quantify the uncertainty of prediction, provides the RUL prediction with the prediction interval, and has higher practical value. The method and the device for predicting the residual service life based on the system operation data acquired by the multiple sensors can be applied to a health management system of complex equipment, help intelligent maintenance of the equipment in the operation process, and are suitable for equipment with multiple sensor data bases, such as turbine engines, mechanical bearings and the like.
According to the fault prediction method based on the sparse ESGP and the multi-objective optimization, provided by the embodiment of the invention, the residual service life of equipment or key structural parts of the equipment is mainly predicted, the introduction of the sparse GP can be adapted to large-scale data, and the complex model parameters are automatically optimized through the multi-objective optimization, so that a better ESGP prediction model is more efficiently selected, reliable RUL prediction with a prediction interval is provided under the consideration of the prediction uncertainty, valuable information is provided for the predictive maintenance of the equipment, the downtime and the maintenance cost of the equipment can be effectively reduced, the operation efficiency and the safety of the equipment are improved, and the fault prediction method has industrial application value.
The fault prediction device based on sparse ESGP and multi-objective optimization provided by the embodiment of the invention is described next with reference to the attached drawings.
Fig. 6 is a schematic structural diagram of a fault prediction device based on sparse ESGP and multi-objective optimization according to an embodiment of the present invention.
As shown in fig. 6, the fault prediction apparatus 10 based on sparse ESGP and multi-objective optimization includes: acquisition device 100, introduction module 200, and prediction module 300.
The collecting device 100 is configured to collect operation data of the device under test, and extract trend characteristics of the device under test according to the operation data. The introducing module 200 is configured to introduce a sparse gaussian process to improve a gaussian process of the echo state gaussian process, introduce a multi-objective genetic algorithm to perform parameter optimization on a model of the echo state gaussian process, and use the effects of point prediction and interval prediction of the remaining service life as targets of multi-objective optimization. The prediction module 300 is configured to perform final prediction on the remaining service life of the device to be tested through the optimized model according to the trend characteristics, so as to obtain a better prediction result. The device 10 of the embodiment of the invention automatically optimizes the complex model parameters through multi-objective optimization, and selects a better ESGP prediction model more efficiently, thereby effectively reducing the downtime and maintenance cost of equipment, improving the operation efficiency and safety of the equipment, and having industrial application value.
Further, in an embodiment of the present invention, the acquisition module 100 is further configured to acquire multi-dimensional feature data acquired by multiple sensors, select to eliminate redundant features according to statistical indexes, perform feature extraction on each dimension of original features, set the length of a time window as a preset value, and take an average value and a differential value of each time window as extracted features to obtain trend features of the device to be tested.
Further, in one embodiment of the present invention, the apparatus 10 of the embodiment of the present invention further comprises: and a modeling module. The modeling module is used for establishing a mathematical model of the echo state Gaussian process, utilizing the trend characteristics as model input and the residual service life as model output, solving an output weight matrix according to an update equation of the echo state network, reserving the iteratively updated reservoir state, combining the model input and the updated reservoir state to obtain the input of the sparse Gaussian process regression model, and training by taking the true value of the residual service life as the target output of the sparse Gaussian process regression model.
Further, in one embodiment of the present invention, the apparatus 10 of the embodiment of the present invention further comprises: and a calculation module. The calculation module is used for selecting a linear function as a mean function and a square exponential function as a variance function before training, and taking the preprocessed test data as input to test to obtain a prediction result of the residual service life after one iteration.
Further, in an embodiment of the present invention, the introducing module 200 is further configured to use a mean value of the prediction result as a point prediction of the remaining service life, and use the variance for constructing a prediction interval to obtain a metric index of the prediction result, obtain a new comprehensive index according to the metric index of the point prediction and the interval prediction of the prediction result, and construct an optimization objective function in combination with the new comprehensive index, thereby converting the multi-objective problem into a single-objective problem, so as to perform optimization with a genetic algorithm, wherein in an optimization process of the genetic algorithm, a cross validation method with weight is introduced and a previously constructed objective function is combined to calculate fitness, thereby selecting a better chromosome, which is a model parameter to be optimized, by iterative optimization to give an optimized sparse ESGP model, and in the testing stage, the training data and the testing data are simultaneously used as input, the likelihood function is selected as Gaussian likelihood, the inference mode is selected as Gaussian inference, and the testing is carried out based on the trained sparse ESGP model so as to obtain the predicted value of the residual service life and the mean value and the variance of the residual service life.
It should be noted that the foregoing explanation of the embodiment of the fault prediction method based on sparse ESGP and multi-objective optimization is also applicable to the fault prediction apparatus based on sparse ESGP and multi-objective optimization of this embodiment, and details are not repeated here.
According to the fault prediction device based on the sparse ESGP and the multi-objective optimization, provided by the embodiment of the invention, the residual service life of equipment or key structural parts of the equipment is mainly predicted, the introduction of the sparse GP can be adapted to large-scale data, and the complex model parameters are automatically optimized through the multi-objective optimization, so that a better ESGP prediction model is more efficiently selected, reliable RUL prediction with a prediction interval is provided under the consideration of the prediction uncertainty, valuable information is provided for the predictive maintenance of the equipment, the downtime and the maintenance cost of the equipment can be effectively reduced, the operation efficiency and the safety of the equipment are improved, and the fault prediction device has industrial application value.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, no attempt is made to limit the disclosure to the particular embodiments or examples described. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.
Claims (2)
1. A fault prediction method based on sparse ESGP and multi-objective optimization is characterized by comprising the following steps:
step S101, collecting operation data of equipment to be tested, and extracting trend characteristics of the equipment to be tested according to the operation data, wherein the method specifically comprises the following steps:
acquiring expressed multidimensional feature data acquired by a plurality of sensors, and selecting to remove redundant features according to statistical indexes;
extracting features of each dimension of original features, setting the length of a time window as a preset value, and taking an average value and a difference value of each time window as extracted features to obtain trend features of the equipment to be tested;
step S102, introducing a sparse Gaussian process to improve the Gaussian process of the echo state, introducing a multi-objective genetic algorithm to optimize parameters of a model of the Gaussian process of the echo state, and taking the effects of point prediction and interval prediction of the residual service life as the targets of multi-objective optimization, wherein the method specifically comprises the following steps:
establishing a mathematical model of the echo state Gaussian process, using the trend characteristics as model input and the residual service life as model output, solving an output weight matrix according to an update equation of the echo state network, and keeping the iteratively updated reserve pool state;
combining the model input and the updated reserve pool state to obtain the input of a sparse Gaussian process regression model, and training by taking the true value of the residual service life as the target output of the sparse Gaussian process regression model;
before training, selecting a linear function as a mean function and selecting a square exponential function as a variance function;
taking the preprocessed test data as input to test to obtain a prediction result of the residual service life after one iteration;
taking the average value of the prediction result as the point prediction of the residual service life, and using the variance for constructing a prediction interval to obtain a measurement index of the prediction result;
obtaining a new comprehensive index according to the measurement indexes of point prediction and interval prediction in the prediction result, and constructing an optimization objective function by combining the new comprehensive index, so that a multi-objective problem is converted into a single-objective problem, and the optimization is performed by using a genetic algorithm, wherein in the optimization process of the genetic algorithm, a cross validation method with weight is introduced and the fitness is calculated by combining the objective function constructed in the front, so that a better chromosome is selected through iterative optimization, the chromosome is a model parameter to be optimized, and an optimized sparse ESGP model is provided by combining the optimized model parameter;
in the testing stage, training data and testing data are simultaneously used as input, a likelihood function is selected as Gaussian likelihood, an inference mode is selected as Gaussian inference, and the testing is carried out based on the trained sparse ESGP model to obtain a predicted value of the residual service life and the mean value and the variance of the residual service life; and
and S103, finally predicting the residual service life of the equipment to be tested through the optimized model according to the trend characteristics so as to obtain a better prediction result.
2. A fault prediction device based on sparse ESGP and multi-objective optimization is characterized by comprising the following components:
the acquisition device is used for acquiring the operation data of the equipment to be tested and extracting the trend characteristics of the equipment to be tested according to the operation data, and is specifically used for:
acquiring multi-dimensional feature data acquired by a plurality of sensors, selecting to remove redundant features according to statistical indexes, extracting features of each dimension of original features, setting the length of a time window as a preset value, and taking an average value and a difference value of each time window as extracted features to obtain trend features of the equipment to be tested;
the modeling module is used for establishing a mathematical model of the echo state Gaussian process, using the trend characteristics as model input and the residual service life as model output, solving an output weight matrix according to an update equation of the echo state network, reserving an iteratively updated reserve pool state, combining the model input and the updated reserve pool state to obtain the input of a sparse Gaussian process regression model, and training by using a true value of the residual service life as a target output of the sparse Gaussian process regression model;
the calculation module is used for selecting a linear function as a mean function and a square exponential function as a variance function before training, and taking the preprocessed test data as input to test to obtain a prediction result of the residual service life after one iteration;
the introduction module is used for introducing a sparse Gaussian process to improve the Gaussian process of the echo state Gaussian process, introducing a multi-objective genetic algorithm to optimize parameters of a model of the echo state Gaussian process, and taking the effects of point prediction and interval prediction of the residual service life as the targets of multi-objective optimization, and is specifically used for:
taking the mean value of the prediction result as the point prediction of the residual service life, using the variance for constructing a prediction interval to obtain the measurement index of the prediction result, obtaining a new comprehensive index according to the measurement index of the point prediction and the interval prediction of the prediction result, and constructing an optimization objective function by combining the new comprehensive index, thereby converting a multi-objective problem into a single-objective problem so as to carry out optimization by using a genetic algorithm, wherein in the optimization process of the genetic algorithm, a cross validation method with weight is introduced and the fitness is calculated by combining the previously constructed objective function, so that a superior chromosome is selected by iterative optimization, the chromosome is a model parameter to be optimized, an optimized sparse ESGP model is given by combining the optimized model parameter, and training data and test data are simultaneously taken as input in a test stage, selecting a likelihood function as Gaussian likelihood, selecting an inference mode as Gaussian inference, and testing based on the trained sparse ESGP model to obtain a predicted value of the residual service life and a mean value and a variance of the residual service life; and
and the prediction module is used for finally predicting the residual service life of the equipment to be tested through the optimized model according to the trend characteristics so as to obtain a better prediction result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810789980.6A CN108961460B (en) | 2018-07-18 | 2018-07-18 | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810789980.6A CN108961460B (en) | 2018-07-18 | 2018-07-18 | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108961460A CN108961460A (en) | 2018-12-07 |
| CN108961460B true CN108961460B (en) | 2020-05-08 |
Family
ID=64496226
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810789980.6A Active CN108961460B (en) | 2018-07-18 | 2018-07-18 | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108961460B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110046378B (en) * | 2019-02-28 | 2022-09-13 | 昆明理工大学 | Selective hierarchical integration Gaussian process regression soft measurement modeling method based on evolutionary multi-objective optimization |
| CN110533251B (en) * | 2019-09-03 | 2020-07-31 | 北京天泽智云科技有限公司 | Method and device for improving adaptive capacity of predictive maintenance model |
| CN110728310B (en) * | 2019-09-27 | 2023-09-01 | 聚时科技(上海)有限公司 | Target detection model fusion method and fusion system based on super-parameter optimization |
| CN113487085A (en) * | 2021-07-06 | 2021-10-08 | 新智数字科技有限公司 | Joint learning framework-based equipment service life prediction method and device, computer equipment and computer-readable storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102736025A (en) * | 2012-06-29 | 2012-10-17 | 沈阳工业大学 | Device and method for predicting electric remaining service life of circuit breaker |
| CN104680024A (en) * | 2015-03-11 | 2015-06-03 | 南京航空航天大学 | Method for predicting remaining useful life of lithium ion battery based on GA (Genetic Algorithms) and ARMA (Auto Regressive and Moving Average) models |
| CN107069122A (en) * | 2017-04-01 | 2017-08-18 | 山东省科学院自动化研究所 | A kind of Forecasting Methodology of electrokinetic cell remaining life |
| CN108009691A (en) * | 2017-12-22 | 2018-05-08 | 东软集团股份有限公司 | Equipment life Forecasting Methodology, device and equipment |
| CN108037463A (en) * | 2017-12-15 | 2018-05-15 | 太原理工大学 | A kind of lithium ion battery life-span prediction method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150254687A1 (en) * | 2014-03-05 | 2015-09-10 | International Business Machines Corporation | Analytics Driven End of Life Product Planning |
-
2018
- 2018-07-18 CN CN201810789980.6A patent/CN108961460B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102736025A (en) * | 2012-06-29 | 2012-10-17 | 沈阳工业大学 | Device and method for predicting electric remaining service life of circuit breaker |
| CN104680024A (en) * | 2015-03-11 | 2015-06-03 | 南京航空航天大学 | Method for predicting remaining useful life of lithium ion battery based on GA (Genetic Algorithms) and ARMA (Auto Regressive and Moving Average) models |
| CN107069122A (en) * | 2017-04-01 | 2017-08-18 | 山东省科学院自动化研究所 | A kind of Forecasting Methodology of electrokinetic cell remaining life |
| CN108037463A (en) * | 2017-12-15 | 2018-05-15 | 太原理工大学 | A kind of lithium ion battery life-span prediction method |
| CN108009691A (en) * | 2017-12-22 | 2018-05-08 | 东软集团股份有限公司 | Equipment life Forecasting Methodology, device and equipment |
Non-Patent Citations (3)
| Title |
|---|
| A Gaussian Process Echo State Networks Model for Time Series Forecasting;Y.Liu;《2013 Joint IFSA World Congress and NAFIPS Annual Meeting》;20131231;正文1-6页 * |
| DesignofNonlinearPredictiveControlforPneumaticMuscleActuatorBasedonEchoStateGaussianProcess;Yu Cao;《sciencedirect》;20171231;全文 * |
| Echo State Gaussian Process;Sotirios P. Chatzis;《IEEE》;20110930;全文 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108961460A (en) | 2018-12-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108961460B (en) | Fault prediction method and device based on sparse ESGP (Enterprise service gateway) and multi-objective optimization | |
| CN109145373B (en) | Residual life prediction method and device based on improved ESGP and prediction interval | |
| CN111178553A (en) | Industrial equipment health trend analysis method and system based on ARIMA and LSTM algorithms | |
| CN114357594B (en) | Bridge abnormity monitoring method, system, equipment and storage medium based on SCA-GRU | |
| CN113723007A (en) | Mechanical equipment residual life prediction method based on DRSN and sparrow search optimization BilSTM | |
| CN112434390B (en) | PCA-LSTM bearing residual life prediction method based on multi-layer grid search | |
| CN117349797B (en) | Aircraft fault detection method and system based on artificial intelligence | |
| CN112347571A (en) | Rolling bearing residual life prediction method considering model and data uncertainty | |
| CN115238850A (en) | Mountain slope displacement prediction method based on MI-GRA and improved PSO-LSTM | |
| CN115902642A (en) | Battery state of charge estimation method and device, electronic equipment and storage medium | |
| CN111783242A (en) | RVM-KF-based rolling bearing residual life prediction method and device | |
| CN114166509A (en) | Motor bearing fault prediction method | |
| CN111400964B (en) | Fault occurrence time prediction method and device | |
| CN112527547A (en) | Mechanical intelligent fault prediction method based on automatic convolution neural network | |
| CN114720129B (en) | Rolling bearing residual life prediction method and system based on bidirectional GRU | |
| CN115292820A (en) | Method for predicting residual service life of urban rail train bearing | |
| CN113743461B (en) | Unmanned aerial vehicle cluster health degree assessment method and device | |
| CN113151842B (en) | Method and device for determining conversion efficiency of wind-solar complementary water electrolysis hydrogen production | |
| CN113159395A (en) | Deep learning-based sewage treatment plant water inflow prediction method and system | |
| CN112183814A (en) | Short-term wind speed prediction method | |
| CN111126694A (en) | Time series data prediction method, system, medium and device | |
| Al‐Dahidi et al. | A novel approach for remaining useful life prediction of high‐reliability equipment based on long short‐term memory and multi‐head self‐attention mechanism | |
| CN116400675B (en) | Fault diagnosis system and method based on improved CNN-LSTM model | |
| CN117155806A (en) | Communication base station flow prediction method and device | |
| CN117521512A (en) | Bearing residual service life prediction method based on multi-scale Bayesian convolution transducer model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |